2013. ‘Negation and polarity.’ In: M. Den Dikken (ed). The Cambridge handbook of generative...

21

Negation and negativepolarity

Hedde Zeijlstra

21.1 Introduction

A universal property of natural language is that every language is able to

express negation, i.e., every language has some device at its disposal to

reverse the truth value of the propositional content of a sentence.

However, languages may differ to quite a large extent as to how they

express this negation. Not only do languages vary with respect to the

form of negative elements, but the position of negative elements is also

subject to cross-linguistic variation. Moreover, languages also differ in

terms of the number of manifestations of negative morphemes: in some

languages negation is realized by a single word or morpheme, in other

languages by multiple morphemes.

The syntax of negation is indissolubly connected to the phenomenon of

(negative) polarity. In short, and leaving the formal discussion for later,

negative polarity items (NPIs) are items whose distribution is limited to a

number of contexts, which in some sense all count as negative. NPIs sur-

face in various kinds of environments and may also vary in terms of the

restrictions they impose on their licensing contexts and the type of licens-

ing relation. Therefore, studying NPIs provides more insight not only into

the nature of such context-sensitive elements, but also into the syntax of

negation itself.

Finally, it should be mentioned that the distinction between negative

elements and NPIs is not always that clear-cut. Inmany languages negative

indefinites, quite often referred to as n-words (after Laka 1990) appear to

be semantically negative in certain constructions, while exhibiting NPI-

like behavior in other configurations. The samemay also apply to negative

markers in some languages.

This chapter aims at providing an overview of themost important recent

findings and insights gained in the study of the syntax of negation and

polarity. Section 21.2 deals with the syntax of negative markers;

Section 21.3 discusses the syntax and semantics of (negative) polarity

items. Section 21.4 will focus specifically on negative concord (i.e., the

phenomenonwheremultiple instances ofmorphosyntactic negation yield

only one semantic negation), with special emphasis on the ambivalent

nature of n-words. Section 21.5, finally, concludes.

One final note: the syntax of negation is complex and touches upon a

multitude of phenomena,many ofwhich cannot be discussed here in detail,

or even at all, for reasons of space. Absence of discussion of such contribu-

tions should not be viewed as an indicator of their (lack of) importance.

21.2 The syntax of sentential negation

In this section I provide a brief overview of the range of variation that the

expression of (sentential) negation cross-linguistically exhibits and what

its underlying syntax is. First, in Section 21.2.1, I introduce the distinction

between sentential and constituent negation, after which I continue by

describing the range of variation that is cross-linguistically attested with

respect to the expression of sentential negation (Section 21.2.2).

Section 21.2.3 deals with the syntactic status of negative markers and,

finally, in Section 21.2.4 their syntactic position is discussed.

21.2.1 Sentential and constituent negationBefore discussing the various ways in which sentences can be made neg-

ative, one important distinction needs to be made. Take for instance the

following minimal pair, dating back to at least Jackendoff (1972).

(1) a. with no clothes is Sue attractive

b. with no clothes Sue is attractive

Although both cases involve the same negative constituent (with no clothes),

(1a) and (1b) crucially differ in their readings. Whereas (1a) denies Sue’s

attractiveness, (1b) entails it, albeit under special circumstances. Also,

syntactically (1a) and (1b) are different, in the sense that (1a) triggers verbal

movement to C0, whereas (1b) does not. Since in (1a) the entire sentence is

felt to be negative, and in (1b) only the PP with no clothes, it is said that (1a)

constitutes an instance of sentential negation, whereas (1b) exhibits con-

stituent negation.

Klima (1964) was the first to offer a number of diagnostics for sentential

negation, such as (among others) continuations by positive question tags

or either phrases; sentences involving constituent negation, by contrast,

can only be followed by negative question tags or too phrases.

(2) a. With no clothes is Sue attractive, is/*isn’t she?

b. With no clothes Sue is attractive, isn’t/*is she?

794 H E D D E Z E I J L S T R A

(3) a. with no clothes is Sue attractive, and/or Mary either/*too

b. with no clothes Sue is attractive, and/or Mary too/*either

Klima’s tests have given rise to a number of criticisms. These criticisms

initially concerned the diagnostics, though not the distinction between

sentential and constituent negation itself. First, the criteria are language-

specific and therefore do not naturally extend to other languages; second,

the Klima tests also take semi-negative adverbs, such as seldom or hardly,

to induce sentential negation (see (4)), even though such elements do

not reverse the polarity of the sentence: (4) does not deny that John drives

a car.

(4) John seldom drives a car, does he?

Finally, the allegedmutual exclusiveness between sentential negation and

constituent negation was called into question. Take for instance (5):

(5) Not every professor came to the party, did they?

Not every professor clearly forms a constituent (a negative DP). Although

examples like (5) are often analyzed as constituent negation (cf. Payne

1985, Cirillo 2009), the diagnostics point in the direction of sentential

negation. It is, thus, a question whether exhibiting constituent negation

is actually incompatible with expressing sentential negation. Rather, what

seems to be the case is that sentential negation should be considered a

scopal notion, rather than a notion in terms of syntactic structure. Then,

(5) is simply an instance of constituent negation that is also able to express

sentential negation.

Following a research tradition that essentially goes back to Jackendoff

(1969, 1972), Lasnik (1975) and many others, Acquaviva (1997) argues

that the notion of sentential negation should be defined in semantic

rather than syntactic terms. Specifically, Acquaviva argues that senten-

tial negation is the result of negating the quantifier that binds the event

variable. In terms of neo-Davidsonian event semantics (Davidson 1967,

Parsons 1990), representations of sentential negation must be repre-

sented as in (6):

(6) John didn’t drive

¬∃e[drive(e) & Agent(j, e)]

Currently, most scholars treat sentential negation a la Acquaviva

(cf. Herburger 2001, Zeijlstra 2004, Penka 2007). Note, though, that adopt-

ing this kind of perspective on sentential negation does not necessarily

preclude the validity of syntactic approaches to the analysis of sentential

negation, as it is generally assumed that existential closure of the predicate

containing the event variable takes place at the level of the vP boundary

(cf. Diesing 1992, Ladusaw 1992, Herburger 2001, Zeijlstra 2004, 2008,

Penka 2007).

Negation and negative polarity 795

21.2.2 Ways of expressing sentential negationThe distinction between sentential and constituent negation paves the

way for one of the central questions that this chapter is about: what are

the syntactic properties of the expression of sentential negation?

Languages exhibit a fair amount of cross-linguistic variation with

respect to the way sentential negation is expressed. However, closer

inspection reveals some remarkable correspondences as well. Let me dis-

cuss two of them.

First, as has been noted by Horn (1989) in his seminal work on negation,

the expression of a negative sentence is always marked in comparison to

its affirmative counterpart. There is no language in the world in which

affirmative sentences are marked and negative ones are not (see also Dahl

1979, Payne 1985). In this respect negative and affirmative sentences in

natural language are not symmetric but rather asymmetric in nature (for a

discussion on this asymmetric view on the positive–negative distinction,

see also Ladusaw 1996).

Second, various strategies for expressing negation turn out to be univer-

sally absent. For instance, no language in the world is able to express

negation solely by means of word order shift, a strategy that is often

exploited to express other grammatical functions, such as interrogatives

(cf. Horn 1989; Zeijlstra 2009).1

This leaves open a syntactically limited set of possible expression strat-

egies: sentential negation must be expressed overtly (i.e., it cannot be left

unspecified), and marking cannot occur merely by way of a word order

shift. This means that every instance of sentential negation must be

expressed by some negatively marked, overt element, with variation

lying only in the type, position, and number of such markers.

Elaborating on Zanuttini’s (2001) state-of-the-art overview, three major

classes of negative elements expressing sentential negation can be

identified.

The first class of strategies concerns negative verbs. In languages like

Evenki (a Tungusic language spoken in Eastern Siberia) special auxiliaries

can negate a sentence. Alternatively, in many Polynesian languages (e.g.,

Tongan) negative verbs even select an entire clause (in a way similar to the

English it is not the case that . . . construction). Examples are shown in (7).2,3

(7) a. bi ∂-∂-w dukuwun-ma duku-ra (Evenki)

I neg -past -1sg letter-obj write-part‘I didn’t write a letter’

b. na’e ‘ikai [CP ke ‘alu ‘a Siale] (Tongan)

Asp neg [asp go abs Charlie]

‘Charlie didn’t go’

The second class of expression strategies is constituted by languages that

make use of negative markers that participate in the verbal inflectional


morphology. An example is Turkish, where sentential negation is

expressed by means of a negative morpheme me that is located

between the verbal stem and the temporal and personal inflectional

afffixes.

(8) John elmalari sermedi (Turkish)4

John apples like.neg .past.3sg‘John doesn’t like apples’

The final class of expression strategies exploits negative particles to

express sentential negation. Negative particles come about in different

forms. Following Zanuttini (1997, 2001) and Zeijlstra (2007), one can dis-

tinguish the following two kinds of negative particles: negative markers

that attach to the finite verb and those that do not.

The first type of these negative particles are negativemarkers that when

expressing sentential negation must be attached to the finite verb.5 Czech

ne and Italian non are two examples:

(9) a. Milan nevola (Czech)

Milan neg .calls‘Milan doesn’t call’

b. Gianni non ha telefonato (Italian)

Gianni neg has called

‘Gianni didn’t call’

In both examples the negative marker shows up in a position to the

immediate left of Vfin. It must be noted though that these markers exhibit

different phonological behavior. Italian non is a separate morphological

word, which for syntactic reasons precedes the finite verb, whereas in

Czech the negative marker is also phonologically attached to Vfin. The

examples above thus show that this first class of these negative particles

is not homogeneous.6

The second class of negative particles is characterized by the fact that, in

contrast with the first class, their syntactic position does not depend on

the surface position of the (finite) verb. Movement of the finite verb does

not trigger displacement of the negative marker. In this respect, the

distributional position of these negative markers is similar to that of

aspectual adverbs, as is shown for German nicht ‘not’ and oft ‘often’ in

(10) and (11).

(10) a. Hans kommt nicht (German)

Hans comes neg‘Hans doesn’t come’

b. . . . dass Hans nicht kommt

. . . that Hans neg comes

‘. . . that Hans doesn’t come’


(11) a. Hans kommt oft (German)

Hans comes often

‘Hans often comes’

b. . . . dass Hans oft kommt

. . . that Hans often comes

‘. . . that Hans often comes’

A final remark needs to be made about the occurrence of multiple

negative markers. Many languages allow more than one negative marker

to appear in negative clauses. Catalan for example has, apart from its

preverbal negative particle no, the possibility of including a second addi-

tional negative particle pas in negative expressions. In Standard French

the preverbal negative particle ne must even be accompanied by the

negative particle pas.7 In West Flemish, finally, the negative particle nie

may optionally be joined by a negative particle en that attaches to the

finite verb (12).8

(12) a. no sera (pas) facil (Catalan)

neg be.fut.3sg neg easy

‘it won’t be easy’

b. Jean ne mange pas (French)

Jean neg eats neg‘Jean doesn’t eat’

c. Valere (en) klaapt nie (West Flemish)9

Valere neg talks neg‘Valere doesn’t talk’

Jespersen (1917) had already observed that examples like the ones in (12)

reflect a widespread diachronic development of languages. Languages like

English, Dutch, Latin, and many others all changed from languages with

only a clitic-like negativemarker through intermediate stages as in (12a–c)

to a stage in which negation is expressed only by means of a postverbal

negative marker. This process is known as Jespersen’s Cycle (after Dahl

1979) and has been formulated by Jespersen as follows:

The history of negative expressions in various languagesmakes us witness

the following curious fluctuation; the original negative adverb is first

weakened, then found insufficient and therefore strengthened, generally

through some additional word, and in its turn may be felt as the negative

proper andmay then in course of time be subject to the same development

as the original word. (Jespersen 1917: 4)

A number of analyses have been presented to account for the range of

variation that is attested cross-linguistically (both synchronically and dia-

chronically) with respect to the expression of sentential negation, of


which a number will be discussed in the next section. However, it must be

noted that this range of variation is not unique to negation. It shows close

resemblance to, for instance, the range of variation that tense, aspect, and

mood markers exhibit, as well as their similar diachronic developments

(cf. Hopper and Traugott 1993, Roberts and Roussou 2003, van Gelderen

2009).

21.2.3 On the syntactic status of negative markersThe question now arises as to what is the exact syntactic status of the

different types of negative particles that have been discussed above, and to

what extent they can be analyzed in formal syntactic terms.

Pollock (1989), based on an intensive study of the distinction between

French auxiliaries and lexical verbs, argues that negative particles, such as

French ne and pas, are base-generated in a particular functional projection,

dubbed NegP, that intervenes between TP and AgrsP. The finite verb, on its

way to T0, then picks up the negative marker ne leaving pas behind in its

specifier position.

(13) [TP Jean ne-mange [NegP pas ne-mange [AgrP mange [VP mange ]]]]

The idea that negativemarkers are hosted in some functional projection in

the clausal spine has strongly shaped the study of the syntactic status of

negative markers, the primary question being which particles may head

such a negative phrase and which ones may not.10

Zanuttini (1997, 2001) already applies a number of diagnostics to prove

that those markers that always show up in the proximity of the finite verb

are syntactic heads that have the entire vP in their complement. One such

diagnostic concerns clitic climbing. In (14b) it can be seen that the pres-

ence of the French negative marker ne blocks movement of the clitic la

from a position within an infinitival complement of a causative verb to a

position adjoining the matrix auxiliary. The example in (14c) makes clear

that this blocking effect is due to the intervening clitic-like negative

marker ne, as clitic movement over pas is not illicit. Zanuttini follows

Kayne (1989b) in arguing that this must be due to ne being an intervening

head blocking antecedent government of the trace, although this analysis

does not crucially rely on Kayne’s explanation, as in other frameworks

intervening heads are also taken to interfere with clitic movement as well

(see Pollock 1989, Travis 1984).

(14) a. Jean la1 fait manger t1 a Paul (French)11

Jean it makes eat to Paul

‘Jean makes Paul eat it’

b. *Jean l1’a fait ne pas manger t1 a l’enfant

Jean it.has made neg neg eat to the child

‘Jean has made the child not eat it’


c. Jean ne l1’a pas fait manger t1 a Paul

Jean neg it.has neg made eat to Paul

‘Jean hasn’t made Paul eat it’

Another diagnostic, also presented in Zanuttini (1997), concerns blocking

of verb movement. Paduan, an Italian dialect from Veneto, requires the C0

head to be overtly filled in yes/no interrogatives.12 In positive interrogatives,

the verb moves from V0 to C0. As a consequence of the Head Movement

Constraint (Travis 1984), such movement would be illicit if another overtly

filled head intervened. Hence, if the Paduan negative marker no is an inter-

vening head, V-to-C movement is predicted to be excluded in Paduan yes/no

interrogatives. This prediction is indeed borne out, as shown in (15).

(15) a. Vien-lo? (Paduan)

comes-he

‘Is he coming?’

b. *Vien-lo no

comes-he neg?‘Isn’t he coming?’

Zanuttini’s analysis that those negative particles that attach to the finite

verb must be heads of some functional projection in the clausal spine is

further proved byMerchant (2006), who developed another diagnostic: the

so-called why not test. Merchant argues that the English why not construc-

tion must be analyzed as a form of phrasal adjunction and therefore it is

predicted that this construction is only allowed in those languages in

which the negative marker is phrasal as well.

(16) [YP [XP why] [YP not]]

As Merchant shows, this prediction is borne out for many of the languages

with a negative particle, illustrated by examples from Italian and Greek

in (17), where the negative particle heads NegP and thus cannot participate

in ‘why not’ constructions:

(17) a. *Perche non? (Italian)

b. *Giati dhen? (Greek)

why neg‘Why not?’

In those languages, in order to express something meaning ‘why not,’ the

construction ‘why no’ (‘no’ as in ‘yes/no’) must be used:

(18) a. Perche no? (Italian)

b. Giati oxi? (Greek)

why no

This observation holds for all languageswhere the negativemarker itself is

not taken to be phrasal, except for those languages where the negative


marker is phonologically identical to the word for ‘no,’ such as Spanish

and Czech.

(19) a. ¿Porque no? (Spanish)

b. Proc ne? (Czech)

why neg /no

The three diagnostics just discussed all show that those negative par-

ticles that attach to the finite verb must be taken to be syntactic heads

within the clausal spine. It is only natural, then, to assume that those

negative particles whose sentential position is in principle independent

of the surface position of the verb should be taken as phrasal elements, i.e.,

not as elements occupying a head position in the clausal spine (leaving

open the question whether these elements are then specifiers of NegP or

not). This assumption indeed appears to be correct.

If negative adverbs are XPs they should not block head movement and

‘why not’ constructions should be acceptable.13 Both predictions are cor-

rect. V2 languages such as Dutch, German, or Swedish exhibit V2 in main

clauses. This implies that the verb has to move over the negative adverb to

C0 in a negative sentence, as is shown for Dutch and Swedish below and

has already been shown for German in (10):

(20) a. . . . om Jan inte kopte boken (Swedish)

. . . that Jan neg bought the book

‘. . . that John didn’t buy the book’

b. Jan kopte inte boken

Jan bought neg the book

‘Jan didn’t buy the book’

(21) a. . . . dat Jan niet liep (Dutch)

. . . that Jan neg walked

. . . ‘that Jan didn’t walk’

b. Jan liep niet

Jan walked neg‘Jan didn’t walk’

From these results it follows that the negative adverbs in (20)–(21)

behave like maximal projections. It is then also expected that these

elements are allowed to adjoin to ‘why’ in the ‘why not’ constructions.

This expectation is confirmed as well, as shown in (22).

(22) a. Why not? (English)

b. Warum nicht? (German)

c. Waarom niet? (Dutch)

d. Varfor inte? (Swedish)

why neg‘Why not?’


So, to conclude, the distinction between the two types of negative particles

can be naturally reduced to a distinction in syntactic phrasal status.

The next question that arises is whether negativemarkers that are instan-

ces of the verbal morphology, such as the Turkish negative marker me,

which precedes tense, mood, and person affixes and follows reflexive,

causative, or passive affixes, are fundamentally different from markers

that attach to Vfin. Can it be the case that they are both base-generated in

some Neg0 position in the clausal spine and only differ with respect to their

morphophonological properties? This question is not restricted to the realm

of negativemarkers, but concerns the comparison between inflectional and

non-inflectional morphemes in general. Traditionally, inflected verbs have

been considered to be the result of a headmovement processwhere the verb

‘picks up’ its affixes (cf. Baker 1985a, Pollock 1989). In this sense, the under-

lying syntactic structure of sentences with a non-phrasal negative particle

and an inflectional negative marker may be identical.

Such a view (present, for instance, in Pollock 1989) is currently in dispute,

however, casting doubt on the idea that inflectional negative markers are

plain syntactic heads, and has been replaced by either lexicalist accounts,

where lexical items enter the derivation fully inflected (cf. Chomsky 1995c

et seq.), or Distributed Morphology-based approaches where the formal

features in the verbal tree are post-syntactically spelled out as either inflec-

tional morphemes or separate words (cf. Halle and Marantz 1993 and sub-

sequent work). Under lexicalist approaches, inflectional markers must be

different from syntactic heads; other approaches question the idea that

syntactically inflectional markers are fundamentally different from syntac-

tic heads: they are only the result of different mechanisms in the spell-out

process. But even under lexicalist approaches the presence of an inflectional

morpheme is connected to a corresponding syntactic head to which the

inflectional morpheme stands in an Agree relation.

Thus, in principle, nothing forbids a unified treatment of non-

phrasal negative particles and inflectional negative markers (i.e., all

negative markers whose sentential position is dependent on the

position of the finite verb) in terms of elements connected to some

head position in the clausal spine. However, before we can draw any

definite conclusions, detailed discussion will be required of the way

that inflectional markers should be treated syntactically. Such a

discussion exceeds the study of negation and is therefore beyond the

scope of this chapter.

21.2.4 On the syntactic position of negative markersThe fact that negative markers can be heads of a particular functional

projection (dubbed NegP) leads to two further questions: (i) What is the

syntactic position of this NegPwith respect to other functional projections


in the clausal spine? and (ii) Is this negative projection also present in

languages that lack an (overt) negative head or do these phrasal negative

markers occupy specifier/adjunct positions of other projections?

Pollock (1989) proposed that NegP is located below TP and above AgrP,

but the exact position of negation within the clausal spine has been the

subject of quite extensive discussion (cf. Belletti 1990, Laka 1990, Zanuttini

1991, Pollock 1993, Haegeman 1995, amongst many others).

Most of these proposals point out that nothing a priori forces the posi-

tion of the negative projection to be universally fixed. Ouhalla (1991a), for

instance, shows that in Turkish, negative affixes are in between the verb

and tense affixes, whereas in Berber, negation is in the outer layer of verbal

morphology, as is shown in (23).14

(23) a. ur-ad-y-xdel Mohand dudsha (Berber)15

neg .fut .3masc .arrive Mohand tomorrow

‘Mohand will not arrive tomorrow’

b. John elmalar-i ser-me-di (Turkish)

John apples like.neg .past .3sg‘John doesn’t like apples’

Assuming that both inflectional negative markers are hosted at Neg0,

Ouhalla argues that the position occupied by NegP in the clause is subject

to parametric variation along the lines of his NEG Parameter (24), which

puts NegP either on top of TP or on top of VP.16

(24) NEG Parameter

a. NEG selects TP

b. NEG selects VP

According to Ouhalla, the different values of this NEG Parameter are also

reflected by the differences in the expression of sentential negation in

Romance languages and Germanic languages. For him, in Romance lan-

guagesNegPdominates TPwhile it does not do so inGermanic languages.17,18

The idea that the position of NegP is more flexible than initially sug-

gested by Pollock (1989, 1993) was further adopted by Zanuttini (1991,

1997). She claims, much in line with the later cartographic approach

initiated by Rizzi (1997) and Cinque (1999) (see Chapter 12), and based on

various Italian dialect data, that different negative markers in Romance

varieties may occupy different positions in the sentential structure and

that universally at least four different NegPs are available (see also Beninca

2006, Poletto 2000, 2008, Manzini and Savoia 2005, for a discussion of

negation in various Italian dialects):19

(25) [NegP1 [TP1 [NegP2 [TP2 [NegP3 [AspPperf [AspPgen/prog [NegP4 ]]]]]]]]

In Zanuttini’s work, different types of negative markers have different

syntactic and/or semantic properties, such as sensitivity to mood (in


many dialects/languages a different negative marker appears if the sen-

tence displays irrealis mood) or the ability to induce sentential negation

without the support of other negative elements (which Italian non is able

to, but French ne is not).

Zanuttini’s proposal has also met with criticism. Whereas her proposal

is essentially right in arguing that more positions should be available for

negative markers, she does not make it clear why these positions should

have to be part of a universal syntactic template. The fact that the distri-

bution of negation appears to be richer than a fixed NegP position suggests

does not necessarily constitute an argument in favor of an even more fine-

grained fixed structure. It might just as well indicate that the syntactic

distribution is relatively free and only constrained by independently moti-

vated syntactic or semantic restrictions.

This is essentially the argument that Zeijlstra (2004) puts forward. He

argues that the minimal (semantic) requirement for a negative marker to

express sentential negation is that it outscopes vP to ensure that sentential

negation is yielded (see Section 21.2.1), and that this constraint determines

the cross-linguistic range for variation. Similarly, Zeijlstra (2006), follow-

ingHan (2001), argues that negationmay never be interpreted in a position

at least as high as C0 in main clauses (as otherwise negation would out-

scope operators with the illocutionary force of a speech act). These two

assumptions thus require negative markers to occupy a position some-

where in the syntactic middle field without alluding to any syntactic

principle (except one that states that semantic scope reflects syntactic

structure; seeMay 1977). Finally, Zeijlstra argues that semantic differences

between different positions (or types) of negative marker should also

result in different scopal effects, i.e., the syntactic position of a negative

marker is (relatively) free, but if the negative marker is included in differ-

ent positions, different semantic effects are expected to arise.

Zeijlstra’s line of reasoning is in line with a series of approaches put

forward by (amongst others) Ernst (2002), Svenonius (2001c), and Nilsen

(2003), who argue that, generally, the fixed orders of adverbials (see

Chapter 13), arguments, discourse particles, etc. does not reflect a prefab-

ricated syntactic template, but rather results from the fact that alternative

orders would lead to semantic anomaly. Consequently, following the anti-

cartographic nature of these approaches (mostly notably Nilsen 2003),

Zeijlstra argues that, whereas negative head markers must head a NegP

of their own, negative specifiers do not necessarily do so. For languages

like Dutch and German, Zeijlstra assumes that their adverbial negative

markers (niet and nicht, respectively) occupy adjunct positions of vP and he

claims that a negative projection is even lacking in the clausal spine.

Zeijlstra’s more flexible analysis of the sentential locus of negation and

negative markers has been adopted by Penka (2007), Cirillo (2009), and

Breitbarth (2009). Breitbarth (2009), and also Haegeman and Lohndal

(2010), argue that an important consequence of Zeijlstra’s approach is


that only negative markers may occupy a Neg0 position. As obvious as this

may sound, closer inspection reveals that this has serious repercussions

for the analysis of negative markers that cannot express sentential nega-

tion without additional support by another negative marker, as illustrated

in (12b–c) and repeated as (26a–b) below.

(26) a. Jean ne mange pas (Standard French)

Jean neg eats neg‘Jean doesn’t eat’

b. Valere (en) klaapt nie (West Flemish)

Valere neg talks neg‘Valere doesn’t talk’

As Breitbarth, and Haegeman and Lohndahl, observe, West Flemish en is

never able to render a sentence negative by itself.20 It is only optionally

available in sentences that have already beenmade negative by other overt

negative elements. For that reason, en, strictly speaking, cannot be taken

to carry some negative feature, which in turn can project Neg0. Instead,

they argue that en carries a weak polarity feature that constitutes a Polarity

Phrase (PolP). Similar arguments have been proposed for the Afrikaans

sentence-final negativemarker nie (Oosthuizen 1998, Biberauer 2008a) and

French ne (Zeijlstra 2009). For the latter Zeijlstra argued that French ne,

being an element that may only survive in (semi-)negative contexts with-

out contributing any semantic negation, should actually be considered a

plain NPI. This already shows the intricate relationship between negation

and negative polarity, the topic of the next section.

21.3 Negative polarity items

The previous section has illustrated that the expression of sentential

negation is subject to a number of both syntactic and semantic constraints.

However, the syntax of negation is not restricted to the syntax of negative

markers and other negative elements only. As has briefly been touched

upon at the end of the previous section, some elements do not induce

semantic negation by themselves, but at the same time only survive in

contexts that, in one way or another, are negative. Such elements are

generally referred to as negative polarity items (NPIs), although other

names surface as well (e.g., ‘affective items’; see Giannakidou 1999).

The most well-known examples of NPIs are formed by the English any-

series, althoughmanymore can be given, e.g., English yet, need, either, or lift

a finger:

(27) a. we *(didn’t) read any books

b. I have*(n’t) been there yet

c. I need*(n’t) do that


d. I *(didn’t) read the book, and John *(didn’t) either

e. nobody/*somebody lifted a finger

NPI-hood is by no means restricted to English. To the best of my know-

ledge, all languages have some NPIs at their disposal (see also Haspelmath

1997 for a non-exhaustive list of languages that display NPIs) and many

languages exhibit a typology of NPIs, often at least as rich as that of

English.21

As has been pointed out by Giannakidou (1999), the term ‘NPI’ in most

cases is amisnomer, asmost so-called NPIs are licensed in contexts that are

not strictly speaking negative as well, such as restrictive clauses of univer-

sal quantifiers, yes/no-questions, or contexts introduced by at most N con-

structions or semi-negative adverbs, such as hardly.

(28) a. every student who knows anything about linguistics, will join the

event

b. Do you want any cookies?

c. at most three students did any homework

d. John hardly likes any cookies

NPIs have received wide attention by scholars in syntax, semantics, and

pragmatics, and they have constituted a fruitful and popular research area

over the past thirty years. As Ladusaw (1996) points out in his seminal

overview article, the study of the behaviour of NPIs has been dominated by

four research questions: (i) the licenser question; (ii) the licensee question;

(iii) the licensing (relation) question; and (iv) the status question.

The licenser question aims at determining what counts as a proper NPI

licensing context. The licensee question asks why certain elements are

only allowed to occur in particular contexts and what distinguishes them

from polarity-insensitive elements. The licensing (relation) question

addresses what kind of constraints the relation between the NPI licenser

and its licensee is sensitive to. Finally, the status question addresses the

status of sentences containing unlicensed NPIs: Are such sentences bad for

syntactic semantic and/or pragmatic reasons? Note that the status ques-

tion is very tightly connected to the licensee question. If it is, for instance, a

syntactic property of NPIs that they require a higher negative(-like) ele-

ment, then a sentence containing an unlicensed NPI is grammatically ill-

formed; on the other hand, if NPIs come along with a pragmatic effect that

causes them to only be felicitously uttered in negative(-like) contexts,

then, by contrast, a sentence containing an unlicensed NPI may still be

grammatical. The four questions thus reduce to three core questions (see

also Giannakidou 2001, 2010, 2011 who in a slightly differently formu-

lated way also presents these questions as core questions in the study of

polarity).

The study of NPI-hood is, however, not restricted to these questions. At

least two additional phenomena merit discussion as well.


First, some NPIs, such as English any-terms, not only survive in negative

or semi-negative environments, but may also appear in positive contexts

inducing a so-called free-choice reading. An example is given in (29) below.

(29) any student of linguistics should read Syntactic Structures

In (29), any is not interpreted as a plain indefinite but receives an inter-

pretation that at least at first sight behaves more like a universal. The

central question in the study to NPIs that can also induce free-choice

readings, such as any, is whether such elements are lexically ambiguous

between an NPI and a free-choice item, or whether these different inter-

pretational (and distributional) properties follow from one and the same

underlying semantic denotation.22

Another thorny phenomenon concerns the alleged mirror image of NPI-

hood. Some elements, like English some or rather, may only appear in

positive contexts. Take for instance (30):

(30) a. I didn’t drink some wine (∃ > ¬; *¬ > ∃)

b. I am (*not) rather ill

Some cannot survive in anegative context. (30a) only has a readingwhere some

outscopes the negation (‘there is some wine that I didn’t drink’); the reading

where some takes scope under the negativemarker is blocked. Rather, in (30b),

cannot occur in a negative context at all. Elementswith this property, such as

some and rather, are referred to as positive polarity items (PPIs). At the same

time, as we will see, NPI and PPI do not always behave on a par, raising the

question as to whether NPI-hood and PPI-hood are indeed two sides of the

same coin, or actually constitute inherently different phenomena.

This section is set up as follows. In Sections 21.3.1–21.3.3, I address the

three questions that originate from Ladusaw (1996), the licenser question,

the licensee/status question, and the licensing question. In Sections 21.3.4

and 21.3.5 I focus on free choice and PPI-hood, respectively.

21.3.1 The licenser questionAs the examples (27)–(28) have already shown, NPIs are only licensed in

particular contexts, some truly negative, some not. The question thus

arises as to what properties constitute NPI licensing environments.

The first and still one of the most important and influential proposals

that tries to reduce all NPI licensing contexts to one single semantic

property, is Ladusaw’s (1979) proposal, based on Fauconnier (1979), that

all NPI licensers are downward entailing (DE), where DE is defined as

follows:

(31) δ is downward entailing iff ∀X∀Y(X⊆Y)→ ([[δ]](Y) ⊆ [[δ]](X))23

To illustrate what is meant here, let’s look at the examples in (32) and (33).

In (32a) the first sentence entails the second one but not the other way


round (32b). This is due to the fact that the set of red shirts is a subset of the

set of shirts. The entailment goes from a set to its supersets.

(32) a. Mary is wearing a red shirt→ Mary is wearing a shirt

b. Mary is wearing a shirt↛Mary is wearing a red shirt

In DE contexts, entailment relations are reversed. This is shown for the

negative contexts in (33) where the only valid inferences are now from a

set to its subsets.

(33) a. nobody is wearing a red shirt↛ nobody is wearing a shirt

nobody is wearing a shirt→ nobody is wearing a red shirt

b. John is not wearing a red shirt↛ John is not wearing a shirt

John is not wearing a shirt→ John is not wearing a red shirt

However, DE-ness is not restricted to negative contexts. Also, the first (but

not the second) argument of a universal quantifier,24 semi-negatives, such

as few, and at most N constructions are DE and license NPIs.

(34) a. every student went to bed→ every linguistics student went to

bed

b. few people sing→ few people sing loudly

c. at most three students left→ at most three students left early

Although this proposal is to be considered a milestone in the study of

NPIs, it faces several serious problems as well, as has often been

addressed in the literature (see the detailed discussions below for refer-

ences). The three most important ones are the following: (i) not every

NPI is licensed in the same sets of DE contexts; (ii) some NPIs can be

licensed in non-DE contexts as well; and (iii) successful NPI licensing does

not depend only on the logico-semantic properties of the NPI licensing

context.

With respect to (i), it can be observed that some NPIs are subject

to different licensing conditions than others. For instance, whereas

English any-terms seem to be fine in all DE contexts, the Dutch counterpart

to any, i.e., ook maar, is ruled out in DE contexts like niet iedereen (‘not

everybody’):

(35) a. nobody / not everybody ate anythingb. [niemand / *niet iedereen] heeft ook maar iets gegeten (Dutch)

nobody / not everybody has prt prt something eaten‘nobody / not everybody ate anything’

Van der Wouden (1994), elaborating on Zwarts (1995), argues that DE

should be thought of as some layer of a negative hierarchy, where the true

negation (not) forms the highest layer, followed by so-called anti-additive

elements (nobody, nothing, no), followed by the next layer, being DE-ness.25

NPIs, then, differ with respect to which layer of negativity is qualified

to license them. English any is licensed in DE contexts (and thus in

all negative contexts), others only in anti-additive contexts (such as


Dutch ook maar) and some NPIs can only be licensed by the sentential

negative marker.26

Although these observations are all empirically correct, it should be

noted that even this classification should be subject to further modifica-

tion. For instance, Hoeksema (1999) shows that Dutch NPI hoeven cannot

occur in the first argument of a universal quantifier, which is DE but not

anti-additive, even though it can occur in other non-anti-additive DE con-

texts such as weinig (‘few’):

(36) a. *iedereen die hoeft te vertrekken, moet nu opstaan

everybody who needs to leave must now get.up

‘everybody who needs to leave, must get up now’

b. weinig mensen hoeven te vertrekken

few people need to leave

‘few people need to leave’

With regard to (ii), Giannakidou (1997, 1999 et seq.) shows that just as

DE-ness is not always a sufficient condition for NPI licensing, it is not

always a necessary condition for it either. For instance, yes/no-questions

are not DE, even though they license NPIs (see van Rooij 2003) and similar

observations have been made for comparatives and superlatives

(cf. Hendriks 1995, Schwarzschild and Wilkinson 2002, Giannakidou and

Yoon 2010).27 Also, Greek tipota ‘anything’ can be licensed under modals

meaning ‘may’ or ‘want’ or in subjunctive clauses (Giannakidou 1997,

1999, 2000).28 Apparently, DE-ness does not seem to be the weakest layer

of negativity and therefore Giannakidou proposes, following Zwarts

(1995), to further extend the hierarchy of negative contexts by an other

layer of negativity: non-veridicality (defined as in (37)).

(37) A propositional operator F is non-veridical if Fp does not entail or

presuppose that p is true in some individual’s epistemic model (after

Giannakidou 1997, 1999, 2010, 2011).

To clarify this, perhaps in (38a) is a non-veridical operator whereas unfortu-

nately in (38b) is veridical, since a speaker uttering (38a) does not take the

sentence John is ill to be necessarily true, whereas a speaker uttering (38b)

does do so.

(38) a. perhaps John is ill

b. unfortunately John is ill

Non-veridicality can be seen as an additional layer of negativity (even

weaker than DE-ness) andmay account for those cases where NPIs, such as

English any terms, may appear in non-DE contexts.29 However, at the same

time, NPIs like any may not appear in all non-veridical contexts, such as

most modal contexts:

(39) *perhaps John read any books


In order to account for this, Giannakidou alludes to the difference between

licensing and anti-licensing: she argues that Greek tipota ‘anything’ may

occur in all non-veridical contexts, i.e., non-veridical contexts license

tipota; by contrast, English any is said to be banned from all veridical

contexts, i.e., veridicality anti-licenses any. This leaves open the possibility

that any is still banned in some non-veridical contexts.

In connectionwith (iii), we should note that under the Ladusaw–Zwarts–

Giannakidou approach, NPI licensing is only dependent on the logico-

semantic properties of the licensing context. This is, however, not always

the case. For instance, conditionals only allow NPIs under particular prag-

matic conditions as discussed by Heim (1984) and von Fintel (1999).

Linebarger (1980, 1987) and Giannakidou (1999) provide additional exam-

ples where contexts that are clearly non-DE or non-veridical still license

NPIs if they come along with a particular negative implicature, as is shown

below:

(40) exactly four people in the room budged an inch when I asked for

help30

The source of licensing in (40) cannot be reduced to the semantic proper-

ties of its position at LF, but seems to lie in the fact that for the speaker the

number of assistants is smaller than expected/hoped for.

The last example suggests that not only semantic but also pragmatic

conditions apply to NPI licensing.

21.3.2 The licensee questionPerhaps evenmore important than the question as to what licenses an NPI

is the question as to what property anNPI has such that it can only occur in

this particular type of context. It is exactly this question which has domi-

nated the study of NPI licensing over the past ten to fifteen years.31

Two types of approaches have been formulated to address this question.

For some scholars, NPI-hood reduces to some semantic and/or pragmatic

requirement that means that NPIs can only be felicitously uttered in

negative contexts of some sort (DE, anti-additive, or non-veridical). For

others, the answer should lie in syntax, i.e., NPIs come along with some

syntactic feature that forces them to appear in negative environments

only.

21.3.2.1 Semantic and/or pragmatic approachesThe first major contribution in the first direction is the widening +

strengthening account by Kadmon and Landman (1993). Their account

consists of two steps. First, they propose that NPI indefinites, such as

English any terms, differ semantically from plain indefinites in the sense

that NPIs are domain wideners. Such domain-widening indefinites extend

the domain of reference beyond the contextual restrictions that plain


indefinites are subject to. Take (41), which contains Kadmon and

Landman’s original examples:

(41) a. I don’t have potatoes

b. I don’t have any potatoes

Whereas (41a) entails that, in a particular domain, the speaker does not

have potatoes, (41b) suggests that the speaker does not even have a single

old potato in some corner in the kitchen.

The second step in Kadmon and Landman’s line of reasoning is that they

claim that sentences containing NPIs like any must be stronger than

sentences containing a plain indefinite. (41b) is stronger than (41a): the

set of situations where (41b) is true is a clear subset of the set of situations

where (41a) is true, so (41b) entails (41a). The strengthening requirement is

thus met. However, the fact that (41b) is stronger than (41a) is due to the

presence of the negative marker: given that negation is DE, removal of the

negation in the examples in (41) would reverse the entailment relation.

Therefore, without the presence of the negation, a sentence like (41b)

would actually be weaker than the sentence without any. Uttering (41b)

without the negation would thus violate the pragmatic strengthening

condition. This is exactly what, for Kadmon and Landman, rules out sen-

tences containing unlicensed NPIs.

The idea that NPIs come along with widening and strengthening effects,

which are responsible for the fact that they can only be felicitously uttered

in DE contexts, has been adopted and implemented in various ways. Krifka

(1995), for instance, argues that the strengthening condition follows as an

implicature as sentences with a weak reading generally bring along an

implicature that the stronger reading is ruled out. In this respect, he

focuses on elements denoting minimal amounts and explains that espe-

cially those elements are prone to become NPIs.

Lahiri (1998) connects the NPI property to NPI even, arguing that the

underlying structure under NPIs is something like ‘even a(n) N,’ based on

data fromHindi,where theword for even is overtly present in indefiniteNPIs:

(42) a. koii bhii (Hindi)32

one even

‘anybody’

b. koii bhii nahiiN aayaa

one even not came

‘nobody came’

A problem, already acknowledged by Krifka (1995) and also present in

Giannakidou (2011), is that under the Kadmon and Landman approach

NPIs pose strengthening restrictions on the contexts that they can appear

in, without such restrictions being encoded in their lexical representa-

tions. Therefore it remains unclear what enforces the fact that sentences

containing NPIs must be stronger than those with a plain indefinite.


In order to ensure that NPIs are always subject to a strengthening require-

ment, Chierchia (2006) proposes that NPIs are domain wideners that carry

an additional, syntactic feature that requires that NPIs must appear under

the direct scope of an abstract strengthening operator that states that any

stronger scalar alternatives of the sentence containing the NPI are false.

Furthermore, Chierchia argues, along the lines of Kadmon and Landman

and Krifka, that strengthened domain wideners always yield a semantic

contradiction unless they appear in DE contexts. If that is correct, it

immediately follows that NPIs are doomed in any contexts other than DE

ones.33

Giannakidou (1997, 2001, 2010) points out that despite the current

popularity of the widening + strengthen approach outlined above, it still

faces several problems.

First she shows that pragmatic infelicitousness and semantic contra-

dictions are generally not judged as being ungrammatical, as shown in

(43) below (taken from Giannakidou 2011).

(43) a. the king of France is my brother

b. John was born in NY, and he was not born in NY

However, the judgments on unlicensed NPIs are much stronger: speakers

generally feel them to be ungrammatical.

Also problematic, she argues, is that Kadmon–Landman type of analyses

only apply to indefinite NPIs. Althoughmost NPIs are indefinites, not all of

them are. For instance, NPIs like either or need are not. Concerning the

latter, as Iatridou and Zeijlstra (2010) have shown, deontic modal NPIs are

actually always universal and never existential. This suggests that, though

not necessarily on the wrong track, the original approach is insufficient: it

is not the only way to explain why NPIs are banned from certain contexts.

It should be noted, however, that most NPIs denote scalar endpoints,

suggesting that scalarity still underlies NPI-hood.34

The third problem for Giannakidou is that widening + strengthening

approaches do not naturally extend to those cases where NPIs are licensed

in non-veridical, non-DE contexts, such as yes/no-questions.

In order to solve this final problem, Giannakidou (2010, 2011) exploits

another type of account, arguing that evenwithin the domain of indefinite

NPIs a distinction must be drawn between those NPI indefinites that

appear in DE contexts only (and could potentially be analyzed as having

their NPI property derived from their domain widening effects) and those

that do not.

For her, this latter type of NPI is lexically deficient for referentiality. She

assumes that NPIs like Greek kanenas ‘anybody,’ which are fine not only in

downward-entailing contexts, but also in all kinds of other non-veridical

contexts, can only be uttered felicitously when they do not have to refer to

some entity in the real world. Therefore, these elements are expected to

not appear in veridical contexts.


21.3.2.2 Syntactic approachesAlthough currently many scholars assume that the ill-formedness of sen-

tences containing unlicensed NPIs is due to pragmatic and/or semantic

reasons, others have argued that those are ungrammatical as a result of

some syntactic constraint.

The tradition that takes NPIs to come alongwith a syntactic requirement

that they be licensed by a (semi-)negative operator goes back to Klima

(1964), and has been presented in more modern frameworks by Progovac

(1992, 1993, 1994), who takes NPI licensing to be some special instance of

syntactic binding, and by Laka (1990), who relates NPIs to the obligatory

presence of an affective phrase (∑P).Postal (2000), followed by Szabolcsi (2004), introduces a revival of

Klima’s theory and claims that NPIs such as English any underlyingly

carry a negation, suggesting a syntactic representation of any as (44).

(44) any: [D NEG [SOME]]

In a negative sentence containing any, the negation moves out of any to a

higher position where it is realized as an overt negator; in semi-negative

sentences this negation may incorporate in other elements.

Den Dikken (2006a) adopts the essence of Postal’s analysis, but modifies

it in more Minimalist terms by assuming that NPIs carry an uninterpret-

able negative feature that must be checked against a negative head in the

clause.35 Independently, and for different reasons, Neeleman and van der

Koot (2002) and Herburger and Mauck (2007) reached this conclusion as

well.

The main problem for such purely syntactic approaches, however, is

that it is hard to understand why most types of NPIs that are attested

always denote some endpoint of a scale. In principle, if NPI licensing is

an instance of syntactic feature checking, all kinds of elements should be

able to act as NPIs, whereas the distribution ofmost if not all NPIs seems to

be restricted semantically.

Herburger and Mauck (2007) try to overcome this criticism by arguing

that the scalar endpoint property is a necessary, but not a sufficient con-

dition for NPI licensing. For them, it is indeed a pragmatic and/or semantic

property whether some element may be a candidate for becoming an NPI,

but that it is only the presence of some uninterpretable negative feature

that turns an element into an NPI.

21.3.3 The licensing questionFinally, all cases discussed so far show that all NPIs must stand in a

particular relation to their licensers. Ladusaw (1979) suggests that, since

the licensing requirement involves a scopal semantic property, this rela-

tion basically boils down to a scope requirement at LF: all NPIs must be

within the scope of a DE operator at LF.


However, as Ladusaw (1979) notes, this constraint on the licensing

relation may be a necessary but not a sufficient condition. NPIs, generally

speaking, may not precede their licenser, even if this licenser outscopes

the NPI at LF. Hence Ladusaw (1979) argues that the c-command relation

must hold not only at LF, but also at surface structure. This now explains

why (45) (taken from Ladusaw 1996) is ruled out.36

(45) *he read any of the stories to none of the children

But, Linebarger (1980) points out that the NPI licensing relation must be

more severely constrained. Concretely, she claims that NPIs must not only

be outscoped by a DE operator at LF, but no scope-taking element may

intervene between the NPI and its licenser either, a claim dubbed the

Immediate Scope Constraint (ISC). Take the following minimal pair

(again from Ladusaw 1996):

(46) a. Sam didn’t read every child a story ¬>∃>∀; ¬>∀>∃

b. Sam didn’t read every child any story ¬>∃>∀; *¬>∀>∃

Although (46a) is ambiguous between a reading where the existential

scopes over the universal and a reverse reading, this second reading is

ruled out in (46b). This directly follows from the ISC, as the NPI then would

not be directly outscoped by a DE operator.37

In themore recent literature on the NPI licensing relation, two phenom-

ena have further refined our understanding of the licensing relation. Both

phenomena are instances of NPI licensing where an NPI is not directly

outscoped by its licenser at LF.

The first phenomenon concerns the difference between direct and indi-

rect licensing (Linebarger 1980, Giannakidou 1999). As discussed at the

end of Section 21.3.1, NPIs are sometimes fine in non-DE contexts as long

as these contexts introduce some negative implicature. The relevant exam-

ple was (40), repeated as (47) below.

(47) exactly four people in the room budged an inch when I asked for help

Obviously, the well-formedness of (47) does not follow under the above-

sketched ISC analysis.

For Linebarger, examples such as (47) show that NPI licensing actually

takes place indirectly. In short, she states that what is responsible for NPI

licensing is that a sentence containing some NPI gives rise to an implica-

ture that contains a negation directly outscoping this NPI. For sentences

already containing a negation this follows straightforwardly; for other DE

operators this implicature needs to be paraphrased in some way such that

it contains a negation (e.g., few X implies not many X). For (47), the required

negative implicature should contain a paraphrase such as ‘not as many

people as I expected.’ Note that as long as a formal computation procedure

of such implicatures is lacking, this type of account cannot make exact

predictions.38


Giannakidou (1999, 2006a) occupies an intermediate position between

Ladusaw’s and Linebarger’s proposals. She takes NPI licensing to be a

relation that takes place at LF between an NPI and a non-veridical

operator and which is subject to the ISC.39 But she also allows NPI rescu-

ing, where a sentence containingNPI that lacks a non-veridical licenser at

LF may be rescued from ill-formedness, if the sentence still gives rise to a

negative implicature. This mechanism is close to Linebarger’s

account, with the difference that for Linebarger all NPI licensing func-

tions in this way, whereas for Giannakidou it is a secondary mechanism:

Giannakidou thus allows NPI licensing to take place at two distinct

levels.40

The second phenomenon concerns another instance of rescuing and is

known as parasitic licensing (den Dikken 2006a). Take the following

example, from Hoeksema (2007).41

(48) ik hoop niet dat je *(ooit) meer van mening verandert

I hope not that you ever(NPI) anymore(NPI) of opinion change

‘I hope that you will never change your opinion anymore’

Ooit and meer are both NPIs, with the crucial difference that ooit may be

licensed by an extra-clausal negation, but meer may not. However, the

licensing ofmeermay be rescued if the clause containingmeer also contains

properly licensed ooit. In this sense, the licensing ofmeer is parasitic on ooit-

licensing.42

21.3.4 NPIs and free choiceIn this subsection and the following one I address two facts that are closely

related to negative polarity: free choice and positive polarity.

Free-choice (FC) items are elements that express indifference or arbitra-

riness (in some form) with respect to a possible referent. Take the exam-

ples in (49):

(49) a. I’ll have whatever you’ll be having

b. Irgendjemand hat angerufen (German)43

FC.person has called

‘some person called’

c. any cat hunts mice

In all these examples the speaker does not seem to impose any restrictions

on the set of possible referents induced by the free-choice element.

Interestingly, in many languages many words display both NPI and FC

effects, for instance Serbo-Croatian ko bilo ‘anybody,’ Malagasy na inona na

inona ‘(lit.) or what or what, i.e., anything’ (Paul 2005), and English any

(being the most stereotypical example). English any may surface, as exem-

plified above, in (a limited number of) positive contexts, as long as it acts as

a FC item.


Given the above, an immediate question that arises is whether English

any and other elements that may manifest both FC and NPI behavior are

lexically ambiguous or have a single lexical representation.

For Kadmon and Landman (1993), domainwidening is a property of both

FC and NPIs. Just as I don’t have any potatoes has a stronger reading than I

don’t have potatoes, Kadmon and Landman argue that (49c) is stronger than A

cat hunts mice. Whereas the generic with the indefinite allows exceptions,

the FC example is more restrictive in this sense. Since the same effect that

drives NPIs to be subject to certain licensing constraints is responsible for

the FC effects, Kadmon and Landman take FC any and NPI any to be a single

lexical item. On similar grounds, Chierchia (2006), Aloni and van Rooij

(2007), and van Rooij (2008) also opt for a unified analysis of FC andNPI any.

The major problem for analyses that argue for such a unified account is

that FC any often seems to yield a universal rather than an indefinite

reading. This is shown in (50), taken from van Rooij (2008), where both

examples convey that all students in Mary’s class are working on NPIs, not

just some, albeit it that the example containing any, but not the one with

every, comes along with a sort of ‘and that’s not an accident’ implicature.

(50) a. any student in Mary’s class is working on NPIs

b. every student in Mary’s class is working on NPIs

For Dayal (1998, 2004) this has been a motivation to radically break with

the idea that FC any is an indefinite and to analyze it instead as a universal

quantifier. FC and NPI any in her view are homophonous.

A problem (indicated by Giannakidou 2001) with analyses that are based

on a treatment of FC any as a universal quantifier is that the universal

reading of FC any is not always manifest. Clearly, the example below does

not mean ‘pick every card.’

(51) pick any card

Thus for Dayal, it is necessary to derive the indefinite reading of FC any

from an underlying universal semantic representation, and the reader is

referred to Dayal (1998, 2004) for the specifics of her proposal based on the

so-called vagueness requirement, which demands that when uttering FC

elements, the speaker must, in some sense, be ignorant or indifferent

about the truth of the entire utterance.

Dayal’s proposal has been criticized by those who seek a unified treat-

ment of FC and NPI any, but also by those who take FC and NPI any to be

both indefinite but nonetheless not identical. The most elaborated pro-

posal along these lines is Giannakidou (2001), also outlined in

Giannakidou (2011), Giannakidou and Quer (1997), and Giannakidou and

Cheng (2006). Giannakidou builds on the insight that both FC and NPI any

are in some sense restricted to appearing in non-veridical contexts only,

but that FC any is furthermore subject to a condition that bans it from

episodic contexts.


Perhaps the most pressing question in the context of FC items is what

the exact semantic effect is that FC generally contributes. For relevant

discussion of this question in the recent literature, the reader is referred

to Kratzer and Shimoyama (2002), Jayez and Tovena (2005), Menendez-

Benito (2005), and Aloni (2006).

21.3.5 NPIs and PPIsA final phenomenon that needs to be addressed concerns positive polarity

items (PPIs). While English any-terms require some DE or non-veridical

licensing context, PPIs, by contrast, are known to be illicit in negative

contexts.

At least four different types of PPIs have been discussed in the literature.

The first type is represented by the English some series and their counter-

parts in other languages (Jespersen 1917, Baker 1970b, Progovac 1994, van

der Wouden 1994, Giannakidou 1997, 2010, 2011, Haspelmath 1997,

Szabolcsi 2004, amongst many others). The second class consists of high

scale elements, such as rather (cf. Krifka 1995, Israel 1996). The third class

of PPIs contains speaker-oriented adverbs and has been thoroughly dis-

cussed by Nilsen (2003) and Ernst (2009). The final class of PPIs concerns

deontic modals, which obligatorily outscope negation, such as English

must (cf. Israel 1996, Iatridou and Zeijlstra 2009, Homer 2010). For an

overview of all types of PPIs, the reader is referred to van der Wouden

(1994) and Israel (2007); for acquisition studies of the relative scope of PPIs

and negation, see Section 25.6.4.

Each type is exemplified in (52). Note, though, that contrary to most

NPIs, PPIs in negative sentences do not always render a sentence ill-

formed, but rather disambiguate them. Therefore, in (52a) and (52d), the

sentences are not ruled out, but rather the readings with the PPI taking

scope under the negation are excluded.

(52) a. John didn’t see somebody

* ‘John saw nobody’√‘there is somebody John saw’

b. I am (*not) rather ill

c. they (*don’t) possibly like spinach

d. Mary mustn’t leave

* ‘Mary doesn’t have to leave’√‘it’s obligatory that Mary leaves’

What PPIs thus show is that they cannot scope below negation. In that

sense, they appear to be the mirror image of NPIs, and various proposals

have tried to understand the behaviour of PPIs in terms of anti-licensing

(Ladusaw 1979, Progovac 1994, amongst others). On the other hand, it has

recently been claimed by many, most notably by van der Wouden (1994),

Szabolcsi (2004), Ernst (2009), and Giannakidou (2011), that PPIs behave


rather differently from NPIs and therefore should call for a different

theoretical treatment.

Szabolcsi (2004), who pursues Postal’s (2000) idea that NPIs underlyingly

carry some negation or negative feature, proposes that PPIs like some

actually have two underlying negative features. Since two negations cancel

each other out, some can naturally survive in positive sentences/environ-

ments. In negative contexts, though, one negative feature is taken care of by

the presence of an overt licenser, leaving the PPI behind with an unlicensed

negation. Therefore, the PPI in a negative context makes the sentence bad.

Evidence for this analysis comes from the fact, already noted by

Jespersen (1917) and discussed in Baker (1970b), that PPIs, surprisingly,

are fine under the scope of two DE/negative operators, strikingly an envi-

ronment where NPIs are generally licensed as well:

(53) I don’t think that John didn’t call anyone/someone ¬ > ¬ > ∃

These facts indeed indicate that PPIs are not simply the mirror image of

NPIs. The reader is referred to Giannakidou (2011a) for a critical assess-

ment of these facts, although she ultimately reaches the same conclusion

as Szabolcsi, namely that PPI-hood must not be explained as the result of

anti-licensing requirements.

A different line of reasoning is explored by Nilsen (2003). Following

Krifka (1995), Nilsen argues that the pragmatic and semantic effects that

Kadmon and Landman take to be responsible for NPI-hood naturally

extend to PPI-hood.44 This idea is also manifest in Ernst (2009), who,

whilst arguing against Nilsen’s scale-based analysis of PPI-hood, endorses

the idea that the PPI status of speaker-oriented adverbs ultimately

reduces to speaker commitment and is therefore pragmatic/semantic in

nature.

Some recent observations by Iatridou and Zeijlstra (2010) also point in

the direction of a unified source for PPI- and NPI-hood. These authors try to

account for the scopal relations that deonticmodals displaywith respect to

(sentential) negation in terms of negative and positive polarity. They argue

that whereas English have to, can, and may are polarity-neutral, other

universal modals are either PPIs (must) or NPIs (need).45 Interestingly,

though, neither PPIs nor NPIs show up in the domain of existential deontic

modals, suggesting that if in one particular domain NPIs surface, PPIs are

also likely to be found there, and vice versa. This makes it likely to assume

that whatever mechanism is responsible for the occurrence of NPIs in a

certain domain also applies to PPIs.

21.4 Negation or negative polarity: negative concord

Although the distinction between negative elements (as discussed in

Section 21.2) and NPIs (as discussed in Section 21.3) appears to be


straightforward – negative elements are semantically negative, NPIs are

not – it turns out that things are not always that clear. In this section, I

present one such case. Take the following Italian examples:

(54) a. Gianni non ha telefonato (Italian)

Gianni neg has called

‘Gianni didn’t call’

b. nessuno ha telefonato

n-body has called

‘nobody called’

In (54a) the semantic negation is introduced by non. The sentence without

non simply means ‘Gianni called.’ In (54b) nessuno acts like a negative

quantifier, such as English nobody, and thus induces the semantic negation.

However, if the two are combined in a sentence, only one semantic neg-

ation is yielded, whereas from compositional perspective two semantic

negations would be expected:

(55) Gianni *(non) ha telefonato a nessuno

Gianni neg has called to n-body

‘Gianni didn’t call anybody’

The phenomenon where two (or more) negative elements that are able

to express negation in isolation yield only one semantic negation when

combined is called ‘negative concord’ (NC) after Labov (1972), and has been

discussed extensively in the past decades.

NC is exhibited in a large variety of languages. Within the Indo-

European language family almost every variety of the Romance and

Slavic languages and a number of Germanic languages (Afrikaans, West

Flemish, Yiddish, and some Dutch and German dialects) as well as

Albanian and Greek exhibit NC.

NC comes about in different forms. In some languages, for example

Czech, a negativemarker obligatorily accompanies all negative indefinites

(or n-words, as Laka 1990 refers to negative indefinites in NC languages),

regardless of their number and position. Those languages are called Strict

NC languages, following terminology by Giannakidou (1997, 2000). In

other languages, so-called Non-strict NC languages, such as Italian, NC

can only be established between n-words in postverbal position and one

negative element in preverbal position, either an n-word or a negative

marker. Examples are below:

(56) a. Milan *(ne-)vidi nikoho (Czech)

Milan neg .saw n-body

‘Milan doesn’t see anybody’

b. dnes *(ne-)vola nikdo

today neg .calls n-body‘today nobody calls’


c. dnes nikdo *(ne-)vola

today n-body neg .calls‘today nobody calls’

(57) a. Gianni *(non) ha telefonato a nessuno (Italian)

Gianni neg has called to n-body

‘Gianni didn’t call anybody’

b. ieri *(non) ha telefonato nessuno

yesterday neg has called n-body

‘yesterday nobody called’

c. ieri nessuno (*non) ha telefonato (a nessuno)

yesterday n-body neg has called ton-body

‘yesterday nobody called (anybody)’

The reader should note that this typology of NC languages is not exhaus-

tive. In languages like Bavarian and West Flemish NC is allowed to occur,

but it is not obligatory (den Besten 1989a, Haegeman 1995). In French and

Romanian the combination of two n-words gives rise to ambiguity

between an NC reading and a reading with two semantic negations, stand-

ardly referred to as a double negation reading (cf. de Swart and Sag 2002,

Corblin et al. 2004, de Swart 2006, 2010, Falaus 2009).

The central question in the study of NC concerns the apparent violation

of semantic compositionality in examples like (55). How is it possible that

two elements that induce semantic negation when used by themselves

yield only one negation when combined?

In the literature, two approaches have been dominant: (i) the negative

quantifier approach, where every n-word is taken to be semantically neg-

ative and where the missing negation in (55) results from some semantic

absorption mechanism dubbed quantifier resumption; and (ii) the

approach that takes n-words to be semantically non-negative NPI-like

indefinites and where the semantic negation in (54b) is only covertly

present. In the remainder of this section I briefly discuss and evaluate

these two main approaches.46

21.4.1 The negative quantifier approachOne proposal, which takes all negative elements to be semantically neg-

ative, goes back to Zanuttini (1991), Haegeman and Zanuttini (1991, 1996),

and Haegeman (1995) and is further formalized by de Swart and Sag (2002).

According to these scholars, NC readings are the result of a process of

negative absorption, analogous to Wh absorption as proposed by

Higginbotham and May (1981). Take (58).

(58) Who loves who?

This sentence can be interpreted as ‘which pairs <x,y> are love-pairs?’ The

process responsible for this reading is so-called quantifier resumption


(cf. van Benthem 1989, Keenan and Westerstahl 1997), which turns a pair

of quantifiers binding a single variable into a single quantifier binding a

pair of variables. Applying this to negation, a sentence containing two

negative quantifiers (like (59)) should then be able to undergo quantifier

resumption as well and yield a reading ‘There are no pairs <x,y> that are

love-pairs,’ which is an NC reading.

(59) nobody loves nobody

However, (59) does not yield an NC reading. Under this quantifier resump-

tion approach, languages cross-linguistically vary with respect to whether

they allow quantifier resumption to apply to sentences containing more

than one negative quantifier or not. NC languages are then languages

where quantifier resumption must then take place, whereas in languages

that lack NC it should not be allowed.47 Also, under this approach negative

markersmust be considered negative quantifiers as well, otherwise the NC

reading of (55) cannot be accounted for.

The power of the quantifier resumption approach is that it can explain

why two negative elements together may yield an NC reading, thus tackling

the compositionality problem. However, it leaves open the question as to

why certain languages should exhibit NC in thefirst place.Why, for instance,

can the negative marker non not be left out in (55)? For de Swart and Sag

(2002), these questions are independent from the question as towhat seman-

ticmechanismderivesNC readings in thefirst place (see de Swart 2010 for an

answer to these independent questions in terms of Optimality Theory).

21.4.2 The NPI approachThe reader will have noticed that the English translations of the NC

examples all contained NPIs. For instance, the Italian example in (55),

repeated in (60a), has the same semantics as the English one in (60b).

(60) a. Gianni *(non) ha telefonato a nessuno (Italian)

b. Gianni has*(n’t) called anybody

In this example nessuno and anybody share two important properties: (i) they

are interpreted as indefinites; and (ii) they must be licensed by negation (non

and n’t, respectively). The similarities between NPIs and n-words do not end

with these two properties. A third striking parallel betweenNPIs and n-words

is that both can appear in constructions which are DE but not anti-additive.

This is shown in the followingSpanish example taken fromHerburger (2001).

(61) dudo que vayan a encontrar nada (Spanish)

doubt that go to find n-thing

‘I doubt that they will find anything’

Here the verb dudo ‘doubt,’ which is DE (but not anti-additive), is able to

establish an NC relation with nada. Examples such as this form a major


problem for theories that take n-words to be negative quantifiers. Without

adopting additional machinery it is impossible for the verb and the quan-

tifier to undergo resumption together.

Given these strong similarities between NPIs and n-words (or polarity

and NC), and in order to solve problems introduced by (61), several schol-

ars have proposed that n-words are in fact special kinds of NPIs. If the

lexical semantics of elements such as Italian nessuno is actually anybody

instead of nobody, the proper readings in (60) and (61) follow immediately.

However, such an approach faces one immediate problem. If n-words are

semantically non-negative, how can the readings of sentences such as (54b)

(repeated as (62) below), where a single n-word induces semantic negation,

be derived?

(62) nessuno ha telefonato (Italian)

n-body has called

‘nobody called’

In an influential proposal by Ladusaw (1992), n-words are said to differ

from plain NPIs in the sense that they are self-licensing, i.e., if nothing else

licenses n-words, NPIs license themselves. But how is this mechanism of

self-licensing implemented within a particular syntactic framework?

Zeijlstra (2004, 2008), following a proposal by Ladusaw, argues that NC is

an instance of syntactic agreement, where n-words are said to be seman-

tically non-negative indefinites, carrying uninterpretable negative fea-

tures that agree with a semantic negation. In cases of self-licensing the

semantic negation is then left phonologically abstract.

Another approach that is based on Ladusaw’s conjecture is Giannakidou

(2001), where it is argued that self-licensing amounts to ellipsis of the

negative marker. This account, however, applies to Strict NC languages

only.

21.5 Concluding remarks

In this chapter I have tried to sketch the major developments in the study

of the syntax of negation and the study of negative polarity, and the ways

these two phenomena can be taken to be connected. Themajor goal of this

chapter has not been to present how certain problems can be solved, but

rather to demonstrate what the important questions are that have arisen

over the past decades. At the same time, I hope this article reveals that

substantial progress has beenmade over the past ten to fifteen years in the

study of negation and polarity.

Concerning the syntax of negation, the various studies of the syntactic

properties of negative markers (most notably Zanuttini’s analyses of neg-

ative markers in Romance varieties) led to a much better understanding of

what constrains the cross-linguistic variation that languages exhibit with


respect to the expression of sentential negation. At the same time, many

more languages, especially outside the family of Indo-European, need to be

investigated to provide a more complete picture.

In the study of negative polarity, we now not only much better under-

stand which properties constitute NPI licensing contexts, we have also

seen answers to the question of why certain elements are sensitive to

negative polarity in the first place. For instance, the Kadmon and

Landman (1993) style of reasoning, especially in Chierchia’s (2006) imple-

mentation, shows that it is possible to derive NPI-hood in terms of domain

widening and strengthening. Many important questions, however, are still

open. One of them is the question how the NPI property follows for non-

indefinite NPIs, such as English need. Another open question is why certain

NPIs require a different type of licensing environment than other NPIs.

Notes

1. This does not mean that the word order in an affirmative sentence is

always the same as in a negatively marked sentence (see Laka 1990 for

examples from Basque).

2. Data from Payne (1985), cited in Zanuttini (2001).

3. For many more examples of negative auxiliaries, see Miestamo (2005).

4. Example from Ouhalla (1991a), also cited in Zanuttini (2001).

5. This type of negative particle has been referred to by Zanuttini (1997)

amongst others as preverbal negative markers, as these negative

markers generally left-attach to the finite verb.

6. See Zanuttini (1997) for amore fine-grained overview of different kinds

of preverbal negative markers based on a survey of Romance micro-

variation (mostly Northern Italian dialects), including a comparison

between preverbal negative markers and other clitics. See also Poletto

(2008) for a further refinement.

7. In colloquial French, though, this negative marker ne is often dropped.

8. Another well-studied language that exploitsmultiple negativemarkers

to express sentential negation is Tamazight Berber (cf. Ouhalla 1991a,

Ouali 2005).

9. Example taken from Haegeman (1995).

10. In more recent Minimalist work, the head–phrase distinction is no

longer taken to be syntactically primitive, but a pure reflex of featural

projection (a head contains a feature that is still able to project; a

phrase does not); see the discussion of Bare Phrase Structure in

Chapters 2 and 4. This, however, does not affect the diagnostics and

dichotomy discussed in this chapter.

11. Examples (14a–b) are from Kayne (1989b), cited in Zanuttini (2001).

12. Cf. Beninca & Vanelli (1982), Poletto (2000), Poletto and Pollock

(2001).


13. Though it should be noted that these diagnostics are less straightfor-

ward under remnant movement approaches to verb movement (cf.

Nilsen 2003, Muller 2004b, Bentzen 2007).

14. See also Ouali (2005) for a discussion of Berber negation.

15. Example taken from Ouhalla (1991a).

16. Ouhalla uses the term ‘parameter’ in the traditional sense. Note,

though, that it is unclear how such a parameter can be implemented

under the Borer–Chomsky conjecture that takes parametric variation

to be stated in terms of lexical features of functional categories (Borer

1984a, Chomsky 1995c). See also Chapters 4 and 25.

17. See also Laka (1990), who argues that a broader Polarity Phrase (PolP),

which also hosts negative markers, merges in English below IP but

above IP in Romance languages.

18. However, Haegeman (1995) argues that, at least inWest Flemish, NegP

is located above TP.

19. It must be noted, though, that Cinque (1999) excludes negation from

the adverbial hierarchy because of its freer distribution.

20. Except for a small number of fixed constructions, such as ‘k en weet (I en

know ‘I don’t know’) in Ghent Dutch (cf. Haegeman 1995).

21. In the remainder of this chapter, for illustration purposes only, I

mostly focus on English NPIs, though.

22. Note that not every NPI or free-choice element exhibits this schizo-

phrenic behavior, nor is it the case that every free-choice item is an NPI.

23. Definition adopted from van der Wouden (1994).

24. Quantifiers denote relations between two arguments: a nominal and a

verbal argument. For instance, in every student sings, student is the first

argument and sings is the second. NPIs may only occur in the first

argument (as shown in (28a)). In the second argument, which is not

DE, NPIs may not show up: *Every student who knows about linguistics, will

join any event.25. A function f is anti-additive iff f(A ∨ B)⇔ (f(A) ∧ f(B)). E.g., no student is

anti-additive, since no student drinks or smokes is truth-conditionally

equivalent to no student drinks and no student smokes. Not every is not

anti-additive as not everybody drinks and not everybody smokes does not

entail that not everybody drinks or smokes.

26. An example of the latter category would be Dutch idiom voor de poes: zij

is *(niet) voor de poes ‘she’s pretty tough,’ cf. van der Wouden (1994).

27. But see Nilsen (2003) for a discussion of these claims.

28. It is important to stress that not only Greek but many other languages

as well exhibit NPIs that can be licensed in all kinds of non-veridical,

but non-DE contexts; examples are Chinese (Lin 1996), Hindi (Lahiri

1998), Salish (Matthewson 1998), Navajo (Fernald and Perkins 2007),

and Romanian (Falaus 2009).

29. See Zwarts (1995) for a proof that all DE contexts are non-veridical.

30. Example taken from Ladusaw (1996).


31. I follow Ladusaw (1996) in calling this question the licensee question.

Others have referred to it differently, e.g., as the ‘sensitivity question’

(Giannakidou 1997) or the ‘compostionality question’ (Giannakidou

2011a).

32. Example taken from Lahiri (1998).

33. The reason why an unlicensed strengthened domain widener yields a

semantic contradiction is the following. Just as (41b) is stronger than

(41a), (41b) without the negativemarker not (*I have any potatoes) should

be weaker than (41a) without the negative marker (I have potatoes).

Given the presence of the abstract strengthening operator, uttering

*I have any potatoeswould entail that I have potatoes is false. But if it is not

the case that I have potatoes, it also follows that I don’t have any

potatoes. Therefore, unlicensed NPIs (for Chierchia, unlicensed

strengthened domain wideners) yield a semantic contradiction. Since

entailment relations reverse in downward entailing contexts, this

contradiction no longer appears under negation. Therefore (41b)

does not yield a contradiction and is fine.

34. Chierchia (2006 et seq.) goes even further than this and argues that this

is the case for all NPIs.

35. Den Dikken does not claim, though, that NPIs always carry an unin-

terpretable negative feature. For English any, for instance, den Dikken

assumes that it is are lexically ambiguous between NPIs carrying such

a negative feature and ‘plain NPIs’ that just have the property that they

need to be licensed by a DE/non-veridical operator at LF, suggesting

that there are actually two different grammatical types of NPI licens-

ing: syntactic and pragmatic/semantic licensing.

36. However, as has been pointed by Ross (1967a), Linebarger (1980), and

Uribe-Etxebarria (1996), NPIs sometimes appear outside the scope of

their licenser at surface structure, as shown below. The example is

from Linebarger (1980).

(i) a doctor who knew anything about acupuncture wasn’t available

37. Interestingly enough, modals do not count as interveners between

NPIs and their licensers (witness the well-formedness of Nobody may

read any book) (cf. von Fintel and Iatridou 2007).

38. The lack of a formal procedure for implicature computation makes

this type of analysis extremely vulnerable to overgeneralization, as

almost every sentence brings in negative implicatures (cf. Krifka 1995).39. Giannakidou is less committed to Ladusaw’s claim that NPIs must be

outscoped by their licenser. For Greek n-words, for instance,

Giannakidou (2000) assumes that these are universal quantifying NPIs

thatmust (directly and immediately) outscope a non-veridical operator.

40. In order to distinguish between licensing in the broad sense (all types

of NPI licensing) and LF licensing of NPIs, Giannakidou refers to the

former as NPI sanctioning.


41. For more examples, see den Dikken (2002, 2006a) and Postal (2000).

42. Parasitic licensing shows some striking resemblances with secondary

triggering (after Horn 2001), where unlicensed NPIs may be rescued by

virtue of the presence in the local environment of another, properly

licensed NPI.

43. Taken from Kratzer and Shimoyama (2002)

44. Israel (1996), on different grounds reaches a similar conclusion.

45. The observation that Englishmust is a PPI is due to Israel (1996) and has

also been entertained by Homer (2010).

46. Other approaches have been formulated as well. Herburger (2001), for

instance, argues that n-words are lexically ambiguous between nega-

tive quantifiers and NPIs. For a detailed discussion of the existing

accounts of NC, the reader is referred to Zeijlstra (2007, 2008). For a

similar discussion on the relation between negative concord and NPI

licensing, the reader is also referred to Penka and Zeijlstra (2010).

47. De Swart and Sag (2002) argue that evidence for this position comes

from languages like French and Romanian, where sentences with two

negative quantifiers are actually ambiguous and can have an NC and a

non-NC reading. For them, in these languages quantifier resumption

optionally applies.


2013. ‘Negation and polarity.’ In: M. Den Dikken (ed). The Cambridge handbook of generative...

Documents

Transcript of 2013. ‘Negation and polarity.’ In: M. Den Dikken (ed). The Cambridge handbook of generative...